Abstract
This chapter introduces clinical decision making, choosing between different options in patient care, with a focus on decisions that could be supported by personalised treatment approaches for CMHDs. These clinical decisions can include those made during
assessment about the nature of the presenting problem (for example possible diagnosis), deciding which treatments will be most appropriate (treatment selection), and decisions made during care (treatment monitoring). Theoretical models of decision making including Prospect Theory (Kahneman & Tversky, 1979), Expected Utility Theory (Schoemaker, 1982) and Bayesian Reasoning (Richardson, 2007) are considered in relation to clinical decisions, as well as factors that can impact on decision making. The final section of this chapter discusses the potential for decision support tools (DSTs), such as clinical prediction algorithms (e.g. the QRISK, Hippesely-Cox, et al., 2007), to support clinical judgement across healthcare. Researchers have used patient characteristics to develop predictive models of treatment response that could inform personalised treatment selection (e.g. DeRubeis et al, 2014), although these have not been translated into available DSTs at present. Systems to aid treatment monitoring decisions have been developed (e.g. Lambert et al, 2001), and these use sessional outcome measurement data to suggest whether the change in symptoms is indicative of a poor treatment prognosis or not. These systems could be further adapted to provide more patient-centred information on treatment progress, for example by identifying groups of patients who respond differently to particular treatments. Currently, there are no routinely available DSTs that can inform both treatment selection and treatment monitoring decisions in CMHDs, but there is potential for such a system to be developed using data collected routinely by IAPT services.
Introduction
Clinical decision making in healthcare can be defined as choosing between alternative options in the care of a patient (Dowding & Thompson, 2003). This can include decisions about which of the available treatments would be most appropriate, whether to stop or continue with the current treatment, as well as decisions about adjusting treatment as it progresses. These judgements are likely to have an effect on a patient’s wellbeing and therefore a clinician will usually aim to choose the optimal treatment, both to improve the patient’s wellbeing and efficiently use healthcare resources.
How decisions are made can depend both on the complexity of the decisions and the characteristics of the decision maker. For more simple and routine decisions the use of intuition (understanding without the need for conscious reasoning) and heuristics (simple automatic rules used in judgement) may be appropriate, however as decisions become more complex, a more analytical or evidence-based approach may be necessary (Bhugra, 2008). Clinical judgement is developed through training and practice (Kienle & Kiene, 2011), therefore the amount of previous experience with specific clinical situations will contribute to variations in the way that decisions are made. This would suggest that clinicians with less experience will be at higher risk of making incorrect or sub-optimal decisions, which may be evident in audits reporting that less experienced clinicians are associated with higher treatment costs than more experienced colleagues (Mehrotra et al., 2012).
In the UK NHS it is recommended that clinicians adhere to published guidance on evidence- based care, such as the National Institute for Health and Care Excellence (NICE) guidelines, when making clinical decisions across healthcare. The development of treatment guidance is built around the best available evidence to inform and support good clinical decision making and the appropriate choice of treatment (Ioannidis & Lau, 2000). One of the main benefits of evidence-based medicine has been the introduction of more objective and quantifiable estimates of clinical variables into healthcare (Sackett & Rosenberg, 1995), as well as the reduction of uncertainty around clinical decisions, for example due to a lack of previous experience with a particular clinical presentation.
Most healthcare guidelines for UK mental health treatment are diagnosis specific (e.g. depression, social anxiety disorder) and evidence on the effectiveness of treatments is typically gathered from randomised controlled trials (RCTs). RCTs are viewed as the gold standard study design to investigate the efficacy of a given treatment and are therefore critical to the development of the evidence-base about particular treatments. However a common criticism of RCTs is that the participant inclusion criteria of many trials can be restrictive, and so those taking part in such studies may not be fully representative of the population of patients attending routine treatment services (Zimmerman, Mattia, &
Posternak, 2002). As a result, clinicians regularly supplement knowledge gained from clinical guidance by using their own specialist knowledge and previous experience to support clinical decision making (Schwartz & Elstein, 2009). This leaves a clear gap for clinicians hoping to not only deliver evidence-based treatments but to offer their patients the ‘best available’ treatment for them as individuals.
Adopting a personalised medicine approach offers the opportunity for clinicians to provide treatments tailored to the individual presentations of their patients and to offer these treatment options based on sound evidence for the likelihood of achieving a desired clinical outcome. This approach will require clinicians to take into account multiple patient
biomarkers (including genetic information), social and family factors, in addition to diagnostic information (Konrad et al., 2015; Pirmohamed, 2014).
Shared decision making
Although decision making has often been considered a clinician based function, increasing value has been placed on the involvement of the patient in clinical decisions (Spatz, Krumholz, & Moulton, 2016). Shared decision making (SDM) can be defined as a collaborative process where the patient and clinician participate in care decisions jointly, discussing the options, potential harms and benefits as well as patients goals (Hoffmann, Montori, & Del Mar, 2014). SDM is a more inclusive approach to treatment, and can increase a patient’s understanding of the likely treatment outcomes, which may be an important factor in patient consent to treatment. This collaborative approach can empower patients, and can lead to the selection of treatment that increases the likelihood of outcomes prioritised by the patient (Hargraves & Montori, 2014).
However, all patients are different and despite the potential benefits of SDM, research has found that some patients may not want to engage with the process, and instead may prefer to trust clinician judgement (Eliacin, Salyers, Kukla, & Matthias, 2015). Involvement in SDM can also fluctuate over time and the course of treatment. Patients are usually keen to have all the information initially, but are often happy to allow the decisions to be made by the clinician as treatment progresses (Deber, Kraetschmer, & Irvine, 2014). Research into the association between patient preferences and treatment outcomes has reported significant variation regarding the potential benefits. Williams and colleagues (2016) suggested that patients who expressed preferences were less likely to report that treatment had helped them when their preferences had not been met. Other researchers have found that meeting treatment preferences reduced the risk of treatment dropout, but did not significantly affect treatment outcomes (Dunlop et al., 2017). The variation in findings may be linked to the methods used to report treatment outcome, as Eiring et al (2015) found that often outcomes from studies exploring the impact of preference on response to pharmacological treatment used clinician reported rather than patient elicited symptom measures.
Types of clinical decisions
The types of clinical decisions made during the course of assessment and treatment can broadly be grouped as: i) decisions made about possible diagnoses or clinical problems; ii) decisions regarding the most appropriate treatment to prescribe/allocate, and iii) decisions made in response to treatment progress through monitoring. These are briefly described below:
Assessment and diagnosis
Information gathered as part of the initial assessment is fundamental to understanding the presentation and underlying issues causing the patient discomfort or distress. This will likely include clinical symptoms such as low mood or worry, and in mental health settings will regularly involve assessment of the social and occupational impact of these symptoms. This information may be used to diagnose the presenting problem, or to formulate the problem and consider what is maintaining/preventing it from improving without any treatment. If there is insufficient evidence available in the initial assessment then further assessment may be required, such as the use of specialist diagnostic equipment in physical healthcare (e.g. CT scans) or further psychological assessment (e.g. cognitive assessment) in mental health settings.
Treatment selection
Following the assessment of the presenting problem(s) and maintaining factors, the next clinical decision will concern the selection of appropriate treatment. The personalised medicine approach aims to incorporate patient characteristics such as demographics and clinical symptoms into this decision, alongside relevant clinical guidance.
For some clinical scenarios the decision will be straight forward due to either the nature of the condition, or lack of alternative treatment options. However, some clinical decisions, especially those with more complex presentations or scenarios where a number of
alternative options are available, will be more challenging for the decision maker. Identifying patient characteristics that are associated with treatment outcomes could therefore help to determine the most appropriate choice. For example, in the treatment of most breast cancers there are a number of options to be considered, but research has shown that patients with higher expression of human epidermal growth factor (HER-2) respond particularly well to Herceptin (Verma & Mukesh, 2012), which can therefore inform the treatment selection decision. In IAPT services, a personalised treatment approach would aim to help make decisions about which type, or intensity of psychological intervention is the most appropriate given the patient’s characteristics at presentation to the service.
Monitoring treatment
Once the selected treatment has been initiated, there will likely be further decisions made in response to information collected during routine monitoring of the patient, and their progress during treatment. This monitoring will provide information about the impact of the treatment on the patient’s wellbeing, for example whether treatment is having the desired effect in reducing symptoms, or if there are significant side effects which may indicate that a change in treatment should be considered. The use of routine outcome measurement (ROM), the collection of patient information at multiple points during the course of treatment, is usually
has been limited response to treatment, then this can inform decisions about whether the treatment should be continued or if an alternative approach should be considered.
Theoretical models of clinical decision making
Given the fundamental role that decision making plays in healthcare there have been various attempts to theorise about how these decisions are made, although relatively little research has been conducted in the context of mental health treatment (Wills & Holmes- Rovner, 2006). Research has suggested that experienced clinicians will often use initial referral information to form an early hypothesis about diagnoses and or appropriate treatment, before or during their first contact with a patient (Elstein & Schwartz, 2002). Clinicians will most often consider clinical guidance and evidence-based knowledge when making their decisions, but those with less experience may not be as able to supplement this with clinical judgement. Instead, clinical judgement is thought to improve over time with experience and training, until in some situations judgement may become automatic. These automatic decisions have been referred to as ‘affective heuristics’ (Slovic, Finucane, Peters, & MacGregor, 2002), and will develop with experience. Although clinician experience has been associated with increased healthcare costs (Mehrotra et al., 2012), research in psychotherapy outcomes suggests that patient outcomes for more experienced clinicians can be worse than outcomes for less experienced clinicians (Goldberg et al., 2016). It may be that more experienced clinicians are more likely to be allocated more complex and difficult to treat patients, but researchers have also suggested a phenomenon known as ‘therapist drift’, where clinicians do not keep up with the evidence-based compared to more recently trained clinicians (Waller & Turner, 2016), which may reduce the effectiveness of interventions they deliver.
Some prominent theoretical models of clinical decision making are described below:
Bayesian reasoning
As the evidence-based approach to clinical guidance in CMHDs has broadly focused on identifying the best treatments for specific conditions, usually one of the first decisions for a clinician will be a formulation of the patient’s problem in the context of their current
circumstances. This may include the identification of a potential diagnoses, which could then be used to inform further clinical decisions, for example drawing on knowledge of relevant clinical guidance to support a treatment plan. The decision as to whether or not a specific diagnosis is present can be represented as a probabilistic choice between an event (diagnosis) existing or not, as a model of Bayesian reasoning (Richardson, 2007).
Bayesian reasoning is derived from Bayes theorem (Bayes, 1763), and posits that clinicians use information from both the patient assessment and existing knowledge about the
likelihood of the event occurring (pre-test probability) to determine whether a diagnosis is present or not. The pre-test probability is based on the likelihood of that disorder naturally occurring in that clinical environment (i.e. prevalence of the disorder in the service), and therefore relies on either the clinician’s previous experience of that diagnosis in the service. Patient information collected at assessment is then used to increase or decrease the clinician’s estimated likelihood of the diagnosis (Fahey & Van Der Lei, 2009). As new information is acquired during assessment, the clinician’s estimate of the likelihood of diagnosis is updated (Schwartz & Elstein, 2009). The post-test probability of the disorder being the ‘correct’ decision is therefore a function of the pre-test probability and the strength of the available evidence.
Bayesian diagnostic reasoning has generally been applied to physical healthcare, where there is more availability of objective measurement values (e.g. blood pressure), whereas the measurement of symptoms in mental health relies more on the subjective measurement of psychological distress from symptom scales (Bhugra, 2008). However, Bayesian
reasoning could be evident in mental health treatment services where a clinician may combine assessment information with the local prevalence of CMHDs (pre-test probability) then estimate the probability of a specific diagnosis (e.g. panic disorder) being present. If the clinical presentation suggests a number of symptoms common to panic disorder, and there is a sufficient pre-test probability of individuals with panic disorder being referred to the service then the clinician may decide that panic disorder is the likely diagnosis and an appropriate treatment plan can be formulated.
Prototypes
An alternative theory about how clinicians make decisions about both the presentation of the patient and appropriate treatment choices is the use of in-built “prototypes”, representations of particular illness/disorders constructed by the clinician (Garb, 2005). These prototypes are internally derived representations of how a typical ‘type’ of patient, for example with social anxiety disorder, would present to services and the clinician would compare a new patient against their existing prototypes to identify an appropriate match (the prototype which appears most similar).
The use of prototypes to identify stratified groups of patients with similar characteristics complements the aims of personalised medicine approaches to treatment, as certain prototypes may be associated with differential outcomes to treatments in IAPT services. However, as these representations are generated by individual clinicians, they will be highly subjective and heavily biased by previous experience of different patient groups. The use of
diagnostic studies, as different clinicians are likely to have slightly different prototypes for the same diagnosis (Pies, 2007). There could be some overlap between the patient
characteristics common to prototypes developed by two independent clinicians, but a system of grouping patients that is to have utility across services will need a more objective method of stratifying patients that would be common to all clinicians.
Expected Utility Theory
Treatment selection decisions will consider which of the available treatments are most likely to result in the best outcome for the patient, and therefore require some estimate of the potential value of the treatment outcomes to the patient. The most commonly cited
normative theory of clinical decision making is Expected Utility Theory (EUT) (Schoemaker, 1982), which is used across healthcare to model decisions (Chapman & Sonnenberg, 2000). In EUT, each possible outcome from each available treatment is given both a likelihood of occurrence (probability) and a value of that outcome to the patient (utility). Utilities are usually given as a range from 1 to 0, with 1 being a perfect state (perfect health) and 0 the worst state (e.g. death), and therefore the best decision is one that results in the highest utility.
These utilities are normally taken from patient and clinician recorded measures of quality of life, and in the context of CMHD treatment could be linked to the level of decreased distress and functional impairment caused by the clinical symptoms. The most appropriate treatment is therefore the one with the highest probability of the best outcome for that patient. The probability of outcomes is derived from either the clinician’s prior knowledge (experience) or from published research if it exists for the specific situation. Therefore, the clinician’s
decision is based on their subjectively judged probability of whether the event will happen or not. EUT also attempts to explain how clinicians consider trade-offs in clinical decisions by weighing up the benefits and costs of certain decisions on eventual outcome (Wills & Holmers-Rovner, 2006).
Prospect Theory
An alternative theory of decision making in the presence of clinical uncertainty is Prospect Theory (Kahneman & Tversky, 1979; Tversky & Kahneman, 1992) and differs from EUT as it is concerned with describing observed decisions assuming the decision maker has inbuilt descriptive rules, rather than assuming the decision maker is rational and able to estimate perfect accuracy (normative). Prospect theory suggests two phases in decision making; an initial editing phase before a subsequent evaluation stage. In the editing phase, all potential outcomes from the decision are ranked based on certain heuristics of the decision maker, specifically around an outcome they consider a reference point and compared to which all
other outcomes are either losses or gains. Using this method, the probability of events occurring can be compared. In the evaluation phase, values are allocated to each outcome (comparable to utilities in EUT) and the decision is made in light of the outcome with the highest value. The major difference between prospect theory and EUT is that the decision maker in EUT does not compute a reference point and therefore focuses only on improving gains, rather than reducing losses in specific situations. In IAPT services this may be a comparison of the probability of treatment success (utility) for either low intensity or high intensity treatment, or whether there is an increased value of allocating to more resource intensive HI treatment over LI interventions.
The similarities between Bayesian reasoning, Prospect Theory and EUT models of decision making are that they all suggest the decision making will use the expected probability of an event (diagnosis or outcome) to inform the decision. This requires either prior experience of the situation or some reference with which to predict the likelihood of the event. As this information is not always available, decision making in these circumstances is vulnerable to bias resulting in either incorrect or over risk adverse decisions that may not be favourable to the patient.
Challenges in making effective decisions
The models of decision making discussed above (for example EUT) propose that the clinician making the decision requires a reliable estimate of both the presence of a specific disorder and the expected value or benefits of appropriate treatment. However, research has frequently shown that clinician’s estimates are prone to biases and errors, in both test and clinical environments. Research from the US has indicated that of closed malpractice cases across all healthcare settings, 64% involved diagnostic error suggesting that incorrect treatments may have been selected due to incorrect estimates from the clinician (Gandhi et al., 2006). It is possible that there is an increased risk of this ‘diagnostic error’ in mental